LilyErnest
Autotuned kernels
Core trick
Benchmarks candidate kernel configurations for deep learning operations and selects the fastest one automatically.
Tags
Description
Autotuned kernels
README excerpt
# InServiceOfX
Monorepo (single or "mono" repository) for deep learning.
## Creating and starting a uv environment for Python 3 to get started
```
uv venv
source .venv/bin/activate
```
Then proceed to build one of the Docker images in `Scripts/DockerBuilds/Builds/` (Make a local copy and configure the necessary `*.txt` files, then run `python BuildDocker.py`)
## (Deprecating) Creating and starting a virtual environment for Python 3
Create a directory for a virtual environment:
```
/InServiceOfX$ python3 -m venv ./venv/
```
Activate it:
```
/InServiceOfX$ source ./venv/bin/activate
```
You should see the prompt have a prefix `(venv)`.
Deactivate it:
```
deactivate
```
## Setting up a Virtual Machine
Requiring manual effort:
- edit `/etc/docker/daemon.json` for place to put Docker containers, in field "data-root"
* References:
- https://medium.com/@calvineotieno010/change-docker-default-root-data-directory-a1d9271056f4
- https://diegocarrasco.com/change-docker-data-directory-vps-optimization/
*
```
docker info | grep 'Docker Root Dir'
```
Example `/etc/docker/daemon.json`
```
{
"data-root": "/mnt/ey/docker",
"runtimes": {
"nvidia": {
"args": [],
"path": "nvidia-container-runtime"
}
}
}
```
However, read about the difference between storage driver "overlay2" for Docker vs. storage driver "vfs":
https://docs.docker.com/storage/storagedriver/overlayfs-driver/
https://docs.docker.com/storage/storagedriver/v